DeepSeek has announced significant updates to its API, including the launch of two new models: V4-Pro and V4-Flash. These models are accessible via both the OpenAI ChatCompletions interface and the Anthropic interface, using the existing base URL with model parameters set to deepseek-v4-pro or deepseek-v4-flash.
The legacy model names deepseek-chat and deepseek-reasoner will be discontinued on 2026-07-24. During the transition period, these names point to the non-thinking and thinking modes of deepseek-v4-flash, respectively. Users are encouraged to migrate to the new model names.
In addition, DeepSeek has rolled out multiple upgrades to its existing models. Both deepseek-chat and deepseek-reasoner have been upgraded to DeepSeek-V3.2, with deepseek-chat corresponding to non-thinking mode and deepseek-reasoner to thinking mode. A special variant, DeepSeek-V3.2-Speciale, is available via a temporary endpoint (base_url set to https://api.deepseek.com/v3.2_speciale_expires_on_20251215) with the same pricing as V3.2 but without tool call support, until December 15, 2025.
Earlier upgrades include DeepSeek-V3.2-Exp and DeepSeek-V3.1-Terminus, both offering thinking and non-thinking modes. The V3.1 update introduced a hybrid reasoning architecture, improved reasoning efficiency, and enhanced agent capabilities, with benchmark scores of 66.0 on SWE-bench Verified, 54.5 on SWE-bench Multilingual, and 31.3 on Terminal-bench.
DeepSeek also highlighted improvements in the deepseek-reasoner model (upgraded to R1-0528), including significant benchmark gains: AIME 2025 rose from 70.0 to 87.5, GPQA from 71.5 to 81.0, LCB_v6 from 63.5 to 73.3, and Aider from 57.0 to 71.6. The deepseek-chat model (upgraded to V3-0324) showed improvements in MMLU-Pro (75.9 to 81.2), GPQA (59.1 to 68.4), AIME (39.6 to 59.4), and LiveCodeBench (39.2 to 49.2).